CIMSS Hyperspectral IR Sounding Retrieval
(CHISR) Processing & Applications
Jun Li@, Elisabeth Weisz@, Jinlong Li@, Hui Liu#, Timothy J. Schmit&,
Jason Otkin@
and many other CIMSS collaborators
@Cooperative Institute for Meteorological Satellite Studies
University of Wisconsin-Madison, Madison, WI
#NCAR/UCAR, Boulder, CO
&Center for Satellite Applications and Research, NESDIS/NOAA
2009 MUG Meeting
02 – 04 June 2009
Madison, WI
Outlines
 Introduction on hyperspectral IR sounder
 CIMSS Hyperspectral IR Sounding Retrieval
(CHISR) Processing
 Hyperspectral IR sounding products and
applications
 Summary and future perspective
2
Hyperspectral IR Sounder
The Infrared Radiance Spectrum
3
Temperatur
e weighting
functions
Moisture
weighting
functions
Abundant Information Content allows the
high vertical resolution temperature and
moisture profiles with high accuracy !
4
High Spectral Resolution - Global AIRS Data
20-July-2002 Ascending LW_Window
5
Brightness Temperature (K)
Hyperspectral/Ultraspectral Infrared
Measurement Characteristics - continue
Texas
Spikes down Cooling with height
(No inversion)
Spikes up Heating with height
(low-level inversion)
Ontario
GOES
GOES
Wavenumber (cm-1)
Detection of inversions is critical for severe weather
forecasting. Combined with improved low-level
moisture depiction, key ingredients for night-time severe
storm development can be monitored.
6
Why sounding retrievals are
needed?
 Easy to use
 Take advantage of full spectral coverage
 Less data volume
7
AIRS
IASI
CrIS
The goal of CHISR is to
provide a physically based
optimal retrieval algorithm to
simultaneously derive
atmospheric temperature and
moisture profiles, surface
parameters and cloud
parameters from hyperspectral
IR measurements (e.g. from
AIRS, IASI, CrIS) alone at
single FOV resolution.
CIMSS Hyperspectral IR Sounding
Retrieval (CHISR) Processing
 Handle surface IR emissivity
 Handle clouds
 Retrieval algorithm
9
Handling surface IR emissivity
 Emissivity spectrum is expressed by
its eigenvectors (derived from
laboratory measurements)
 Regression retrieval are used as the
first guess
 Simultaneous retrieval of emissivity
spectrum and soundings in physical
iterative approach (Li et al. 2007,
2008)
10
Re-group from
IGBP category:
Forests: Evergreen needle
forests Evergreen broad
forests; Deciduous needle
forests; Deciduous broad
forests; mixed forests;
Shrubs: Opened shrubs;
Closed shrubs;
Savanna: Woody savanna;
Savanna;
Cropland: Cropland; Crop
mosaic;
Snow/Ice: Snow; Ice;
Tundra;
Desert: Desert/Barren;
Global AIRS emissivity map – CIMSS research product
(01 – 08 Jan 2004)
Ecosystem land cover
11
Handling clouds
 Using collocated MODIS cloud mask for
AIRS cloud detection (Li et al. 2004).
 Employ a cloudy radiative transfer model
accounting for cloud absorption and
scattering (Wei et al. 2004)
 Retrieval sounding and cloud parameters
simultaneously (Zhou et al. 2007, Weisz et
al. 2007)
12
AIRS BT spectrum
Whole sounding in broken clouds and above-cloud sounding in thick 13
clouds can be derived
First step: Regression
CLEAR
CLOUDY
Training data set
(SeeBor V5)
Cloudy Training data set
(ice, water)
Radiance calculations
Radiance calculations
SARTA v1.7 (UMBC)
Fast RT cloud model (Wei, Yang)
Additional Predictors
(spres, solzen)
IMAPP RTV Software v1.3
BT and Scanang
Classification
Scanang
Classification
PC regression
C  dXAT (AAT )1
Ice Cloud Regr Coeffs
Water Cloud Regr Coeffs
Clear Regr Coeffs

Regression RTV
X  X tr  CAobs
T, Q, O3, STemp, Emissivity

at single FOV
T, Q, O3, STemp, CTOP, COT
at single FOV
To physical inversion

Second step: Physical Inversion
Cost function for a quasi non-linear case:
J  (y  F(x))T S1 (y  F(x))  (x  x a )Sa1(x  x a )
Newton-Gauss Iteration with regularization parameter :
x i1  x a  (K iT S1K i  Sa1 )1 K iT S1[y  F(x i )  K i (x i  x a )]
x, xa
K
Sa, S
F
current /a priori atmospheric state vector
Jacobian
A priori / measurement covariance matrix
Forward model
Transform to EOF space:
c i1  (K˜ iT S1K˜ i  S˜ a1 )1 K˜ iT S1[y  F(x i )  K˜ ic i ]
c  x˜  x˜ a
x˜  x
K˜  K
S˜ a   T Sa 

eigenvector matrix
Second Step: Physical Inversion - flow chart
Eigenvector matrix 
Matrices Sa and S
Regression
T,Q,O3,ctp,cot,de,emiss
1695 cm-1
Forward model
calculations
1350 cm-1
1210 cm-1
water cloud, De=10,Cot=0.5
Jacobian
calculations
Q Jac
COT Jac
Next iteration
dy  y  yc
dyi  dyi1
yes
Decrease 
no (fail)


Increase 
Physical Inversion
Update Profiles
# fails < f max
# iterations < it max
dy > dycrit
yes
no
Physical RTV results
T,Q,O3,ctp,cot,de,emiss
Hyperspectral/Ultraspectral Infrared
Sounding Performance in Clear Skies
Hyperspectral
Broad band
Hyperspectral
Broad band
Can Achieve Improved Atmospheric Profile Accuracy
17
MODIS 1km images (1705, 1710)
eye
Hurricane Isabel case study
environment
AIRS data used for Hurricane IKE (2008)
study
 September 6-7, 2008 (115oW – 30oW; 0 – 40oN)
 About 10 AIRS granules every day
 Single field-of-view (FOV) temperature and
moisture profiles (13.5 km at nadir) from AIRS are
derived using CHISR
 Clear sky only temperature and moisture
soundings are provided in the assimilation
experiment
19
Retrieved 500mb temperature
(2008.09.06 – Used in assimilation)
(K)
(K)
CIMSS/UW
Clear sky AIRS SFOV temperature retrievals at 500 hPa on 06 September
2008, each pixel provides vertical temperature and moisture soundings.
20
Retrieved 500mb temperature
(2008.09.07 - Used in assimilation)
(K)
(K)
CIMSS/UW
Clear sky AIRS SFOV temperature retrievals at 500 hPa on 07 September
2008, each pixel provides vertical temperature and moisture soundings.
21
Tracks of ensemble mean analysis on Hurricane IKE
CTL run: Assimilate radiosonde, satellite
cloud winds, aircraft data, and surface data.
AIRS
(Li and Liu 2009 - GRL)
Analysis from 06 UTC 6 to 00UTC 8 September 2008
22
Track errors of on Hurricane IKE
AIRS
(Li and Hui 2009 - GRL)
Analysis from 06 UTC 6 to 00UTC 8 September 2008
23
SLP Intensity on Hurricane IKE
AIRS
(Li and Liu 2009 – GRL)
Analysis from 06 UTC 6 to 00UTC 8 September 2008
24
Assimilation Experiments (cont 1)
 4-day ensemble forecasts (16 members) from the
analyses on 00UTC 8 September 2008.
 Track trajectory and hurricane surface central
pressure are compared (every 6-hourly in the
plots).
25
Tracks of 96h forecasts on Hurricane IKE
CTRL run: Assimilate radiosonde, satelliteAIRS
cloud winds, aircraft data, and surface data.
With AIRS
No AIRS
(Li and Liu 2009 – GRL)
Forecasts start at 00 UTC 8 September 2008
26
Track errors of 96h forecasts
No AIRS
With AIRS
(Li and Liu 2009 – GRL)
Forecasts start at 00 UTC 8 September 2008
27
Typhoon Sinlaku (2008)
28
29
700 hPa water vapor mixing ratio (g/kg) (Sinlaku – 2008)
Tracks of ensemble mean analysis on Sinlaku
Analysis from 00 UTC 9 to 00UTC 10 September 2008
30
Tracks of 48h forecasts on Sinlaku
CTRL run: Assimilate radiosonde, satellite
cloud winds, aircraft data, and surface data.
CTRL
CIMSS
Hui Liu (NCAR) and Jun Li (CIMSS)
Forecasts start at 12 UTC 9 September 2008
31
Future Geostationary advanced IR sounder provides 4-D
Temperature, Water Vapor, and Wind Profiling Potential
GIFTS - IHOP simulation 1830z 12 June 02
GOES-8 winds 1655z 12 June 02
Simulated GIFTS winds (left) Vs. GOES current oper winds (right)
32
Summary
 CIMSS has developed an algorithm and processing
package for full spatial resolution sounding retrieval from
hyperspectral IR radiances in both clear and cloudy skies
 Hyperspectral IR sounder provides unprecedented global
atmospheric temperature and moisture profiles that are
critical for weather forecast and other applications
 Temperature and water vapor soundings from AIRS
evidently improve ensemble forecast of hurricane track and
intensity for Hurricane IKE and Typhoon Sinlaku (2008)
 Placing a hyperspectral IR sounder in geostationary orbit
will provide four-dimensional fields of moisture,
temperature and wind with high accuracy, which is critical
for high impact weather nowcasting.
33